Seismic Introduction

Introduction to Seismic Imaging
(Refraction and Reflection Seismology)
Mauricio D. Sacchi
Department of Physics, University of Alberta
2
Chapter 1
Ray Theory
In the preceding chapters we have analyzed the solution of the wave equation in terms of P and S waves. We
have shown that the solution of the wave equation leads to the concept of wavefrons. A wavefrint is the set
of points that at a given time undergo the same motion. Rays are defined as the normal to the wavefronts,
in other words rays describe the direction of propagation of the perturbation.
In the plane wave case, the rays are a family of parallel straight lines, in the spherical wave case, the rays
are straight lines radiating from the center of the spherical wave (the position of the source) (Figure (3.1)).
Rays provide an intuitive scenario to understand wave propagation. We will abandon partial differential
equations and we will concentrate on describing time-space curves which arise from ray theory. For this
porpoise we will use two fundamental principles which you might have already seen in optics: Huygens and
Fermats principles.
The geometry of a wavefront is governed by Huygens principle which states that every point in a
wavefront can be considered a small secondary wavelet which travels outward in every forward direction
t3
t3 t4
t1 t2 t2
Rays
Rays t1
Figure 1.1: Plane waves and spherical waves.
3
4 CHAPTER 1. RAY THEORY
with the velocity of propagation of the medium at that point. In fact, if we know the wavefront at a given
instant we can use this principle to draw the wavefront at a position in a future time.
Huygens principle governs the geometry of wavefronts. Fermats principle governs the geometry of
ray-paths. This principle states that the ray will follow a minimum time path.
1.1 Snells Law
It is possible to use Fermats principle to derive Snells law. Consider a ray leaving a point P (Figure (3.2))
in a medium with velocity v1 , which path will tha ray take to arrive to Q?. Assume the velocity of the
second medium is v2 .
In Figure (3.2), I plotted 3 alternative paths, which is the true one? According to Fermats principle the
seismic ray will travel along a minimum time path. To solve the problem we need to obtain an expression of
T (time) in terms of a suitable variable. In this case, I will parameterize my problem in terms of x (Figure
(3.3)). The travel-time to go from P to Q is given by
P O OQ
T (x) = + (1.1)
v1 v2
where
q
PO = (x xP )2 + yP2 (1.2)
q
OQ = 2 .
(xQ x)2 + yQ (1.3)
The minimum is found by equating to zero the derivative of T (x) with respect to the variable x:
dT
=0 (1.4)
dx
dT 2(x xP ) 2(xQ x)
= =0 (1.5)
dx v1 P O v2 OQ
from the above equation we can recognize the following
(x xP )
sin(i) = (1.6)
PO
(xQ x)
sin(j) = (1.7)
OQ
1.1. SNELLS LAW 5
Medium I , v1
o o o
Medium 2, v2
Q
Figure 1.2: Ray paths connecting two points. Which is the correct path to go from P to Q?
Replacing sin(i) and sin(j) into equation (1.5)
dT 2 2
= sin(i) sin(j) = 0 (1.8)
dx v1 v2
the latter is called Snells law, which I will re-write as follows:
sin(i) sin(j)
= (1.9)
v1 v2
In Figure 3.4, I plotted different ray trajectories for a model like the one given in Figure 3.2. In the numerical
example, I have adopted velocities v1 = 2000m/s and v2 = 3000m/s. The travel-time curve (T (x)) shows a
minimum at x = 1000m. That point corresponds to the minimum time path. This is the path that the ray
will use to go from P to Q. I also plotted, sin(i)/v1 and sin(j)/v2 versus x. In the intersection of these two
curves Snells law is satisfied.
Medium I , v1
Medium 2, v2
Q
P =(xp,zp); Q =(xq,zq); o = (x,0)
Figure 1.3: Ray path connecting P and Q, x is the variable of our problem.
Ray paths Traveltimes

2000 3.6
1000 3.4
0
3.2
t(sec)
z(m)
1000
3
2000
3000 2.8
4000 2.6
2000 1000 0 1000 2000 2000 1000 0 1000 2000
x(m) x(m)
Traveltimes 4
x 10
3.6 6
sin(i) / v1
3.4
4
3.2
t(sec)
3
2
2.8 sin(j) / v2
2.6 0
0 0.5 1 1.5 2000 1000 0 1000 2000
Angle i (rad) x(m)
Figure 1.4: Different rays can connect P and Q. A seismic perturbation will travel via a minimum time
path. In this example the velocity of the upper medium is v1 = 2000m/s, the lower medium velocity is
v2 = 3000m/s.
1.1. SNELLS LAW 7
Ray paths Traveltimes

2000 3.45
1000 3.4
0 3.35
t(sec)
z(m)
1000 3.3
2000 3.25
3000 3.2
4000 3.15
2000 1000 0 1000 2000 2000 1000 0 1000 2000
x(m) x(m)
Traveltimes 4
x 10
3.45 3.5
3 sin(i) / v1
3.4
2.5
3.35
2
t(sec)
3.3
1.5
3.25 sin(j) / v
1 2
3.2 0.5
3.15 0
0 0.5 1 1.5 2000 1000 0 1000 2000
Angle i (rad) x(m)
Figure 1.5: Different rays can connect P and Q. A seismic perturbation will travel via a minimum time
path. In this example , the velocity of the upper medium is v1 = 3000m/s, the lower medium has a velocity
v2 = 2000m/s.
We have analyzed what is called a refracted wave. It is important to note that when a seismic wave reaches
a boundary between contrasting materials, we have also reflected waves. We can use Fermat principle to
compute the minimum travel-time condition for a reflected wave. In fact, part of the incident energy is
refracted and part is reflected (Figure 3.6). The geometry of the ray paths is governed by Snells law, for
the refracted wave we have
sin(i) sin(j)
= (1.10)
v1 v1
and for the reflected wave,
sin(i) sin(i )
= => i = i (1.11)
v1 v1
If the velocity of second half space is greater that the velocity of the upper half space (v2 > v1 ) the angles
satisfy the following condition
if v2 > v1 => j > i . (1.12)
When can adopt i as the variable of our problem, and write
v2
j = asin( sin(i)) (1.13)
v1
It is clear that at some point i = ic (critical angle) the following condition is satisfied
v2
90o = asin( sin(ic )) , (1.14)
v1
which can be written as follows
v1
sin(ic ) = (1.15)
v2
This critical angle ic defines a wave that travels horizontally parallel to the interface. This wave is called a
Head Wave, and has the property that it transmits energy back into the upper medium as it travels along
the interface (Figure 3.7).
1.1. SNELLS LAW 9
P Incident
P Reflected
v1
i i
v2 j
P Refracted
sin(i)/v1 = sin(i)/v1 ==> i =i
sin(i)/v1 = sin(j)/v2
Figure 1.6: A P wave incident on an interface between contrasting materials.
Reflected Wave Direct Wave
ic ic V1 sin(i c ) = V1/V2
i i
V2
Head Wave
Head Wave
ic
Figure 1.7: Rays in a one layer structure over a half space.

1.1.1 T X refraction profiles
In a refraction experiment we attempt to estimate the velocity and depth of layers which are assumed to be
of constant velocity and sub-horizontal. We will start to analyze a very simple scenario: one layer over a
half-space.
A source is placed in position A and receivers distributed on the surface of the Earth (Figure 3.7). We
will compute the travel time for a perturbation originated at A to arrive to a point D a distance x from the
source (x is our variable or source-receiver distance).
First, we will analyze the direct wave:
The total travel-time for a direct wave traveling from A to D is given by
Td (x) = x/v1 . (1.16)
This equation defines a straight line in the T x plane. The slope of the line defines the slowness 1/v1 .
Now, we will concentrate on the head wave. At critical angle i = ic the travel time for the head wave is
given by the time to go from A to D via the path ABCD
AB BC CD
THead (x) = + + . (1.17)
v1 v2 v1
It is clear from the diagram that AB = CD, then
AB BC
THead = 2 + (1.18)
v1 v2
2z x 2 z tan(ic )
THead (x) = + (1.19)
v1 cos(ic ) v2
Using the critical angle condition sin(ic ) = v1 /v2 we can write
THead (x) = x
v2 + 2zcos(i
v1
c)
x (1.20)
= v2 + T1
2zcos(ic)
T1 = . (1.21)
v1
The head wave defines a straight line in T x (Figure 3.9). We can compute the velocity of the layer v1 ,
the velocity of the half-space v2 and z by the following procedure:
1.1. SNELLS LAW 11
x
Source Receiver
A D
z
ic ic
B C
T_Head = AB/V1+BC/V2+CD/V1
Figure 1.8: Geometry of a refraction profile.
1. From T (x) measure the slopes and compute v1 and v2
2. Compute ic using the expression sin(ic ) = v1 /v2
3. Measure t1 and compute z = (v1 t1 )/(2 cos(ic ))
The critical distance xc in Figure 3.9 defines the distance where the head wave will first be recorded on
surface of the earth
xc = 2 z tan(ic ) . (1.22)
1.1.2 T X Refraction curves for a dipping layer
We will examine the case when a single dipping reflector is excited by a seismic perturbation at critical angle.
The problem is defined in Figures 3.10 a and b. We have two possible profiles, the first one corresponds to a
source located in shallower part of the dipping bed, the second to the case when the source in located in the
deeper part of the dipping bed. We will see that the combination of these two experiments will allow us to
estimate the unknowns of the problems. (These experiment are also called Profile and Reverse Profile)
In the preceding section we solved the refraction problem for a non-dipping interface. In that case we
used a single profile and derived v1 , v2 and z. In the dipping layer case we need to estimate another variable
x
Source R1 R2 R3 R4 R5
A D
z
ic ic
B C
(Direct Wave)
Slope=1/V1
(Head Wave)
Slope = 1/V2
Time
t1 t1 = 2 z cos(i )/V1
c
x
x
c
Figure 1.9: Travel-time curve for a refraction profile.
(the dip of the layer).
The total travel-time for the Experiment I (Figure 3.10a) is given by
AB BC CD
TI = + + . (1.23)
v1 v2 v1
But,
z
AB = (1.24)
cos(ic )
Z
CD = (1.25)
cos(ic )
BC = x cos() Z tan(ic ) z tan(ic ) . (1.26)
The time in terms of our variables and unknowns is given by
1 tan(ic ) 1 tan(ic ) x cos()

TI = z [ ]+Z[ ]+ . (1.27)
v1 cos(ic ) v2 v1 cos(ic ) v2 v1
1.1. SNELLS LAW 13
the last equation can be reduced by introducing the following equations
1 sin(ic )
= Critical angle condition (1.28)
v2 v1
and
Z = z + x sin() (1.29)
After substituting the last two equation in (3.27) we obtain
2z x
TI = v1 cos(ic ) + v1 [sin() cos(ic ) + cos() sin(ic )]
2z x (1.30)
= v1 cos(ic ) + v1 sin(ic + ) .
The last expression corresponds to a straight line with slope:
sin(ic + )
slopeI = (1.31)
v1
and intercept time (T (x = 0))
2z
InterceptI = cos(ic ) . (1.32)
v1
The direct ray can be used to obtained v1 .
We can obtain similar expression for the Experiment II (source in D and receiver in A), in this case we
have
2Z x
TII = cos(ic ) + sin(ic ) . (1.33)
v1 v1
which also defines a straight line in T X,
sin(ic )
slopeII = (1.34)
v1
2Z
InterceptII = cos(ic ) . (1.35)
v1
In Figure 3.11, I show the T X curve for the Experiments I and II. This is the procedure to compute our
unknowns:
1. From the direct rays compute v1 (Use both profiles and average the results or use the profile with the
best SNR).
2. The slopeI and slopeII are measured. At this point we have two equations with two unknowns ic and
. You can easily solve the 2 system of equations.
3. Once we have obtained v1 and ic use the critical angle condition to compute v2 , (sin(ic ) = v1 /v2 ).
4. Use the intercept times to compute z and Z.
1.1.3 Fitting straight lines to first arrival picks
In a real experiment we will pick first arrivals and we will construct a table like the following
event : Direct Wave pick x(m) t(msec)

---------------------
1 15 10
2 30 30
3 40 42
event : Refracted wave pick x(m) t(msec)

---------------------
1 100 51
2 120 63
3 150 74
4 185 85
When dealing with real data we rarely have perfect straight lines. In this case, we need a technique to
fit a straight line to our observations.
Suppose that we want to fit a line to the refracted wave. The picks for the refracted wave are given by
xi , ti , i = 1, np (np : number of picks). The analytical expression of the straight line is
t(x) = ax + b . (1.36)
we can use least squares to find a and b. The least squares method finds a solution for a and b based on the
minimization of the error function:
X
J= (ti axi b)2 . (1.37)
i
1.1. SNELLS LAW 15
(a)
Experiment I
A D
ic
z
ic Z Medium I, V1
Medium II, V2
Source : A, Receiver: D
(b)
Experiment II
A D
ic
z
ic Z Medium I, V1
Medium II, V2
Receiver: A, Source:D
Figure 1.10: Refraction experiment. Now we have an extra unknown , we will run two experiments (a) and
(b) to determine z, v1, v2 and .
Refraction Experiment II
Time
Refraction Experiment II
Direct
Direct
Intercep I
Intercept II
Figure 1.11: Refraction experiment. Travel-time curves
The minimum error is found by taking derivatives of J with respect to a and b. The latter leads to a
2 2 system of equations from where we can compute a and b. Once we have a and b we can compute the
analytical line that best fit the observations in the least squares sense.
Sometime least squares is not a good approach to fit curves. This is particular true when our observations
are contaminated by outliers (gross errors). In this case, we need a robust technique like least absolute error
X
l1 = |ti axi b| . (1.38)
i
The minimization of l1 leads to the unknowns of our problem a and b.
In Figure 3.12, I use least-squares to fit the direct and the refracted travel-times to a set of noisy
observations. In Figure 3.13, we analyze the effect of a out-lier in the least-squares fitting. It is clear that a
more reasonable solution is found using the robust fitting (l1 ).
1.1. SNELLS LAW 17
. .
. .
.
. .
Time
Least squares fitting
..
.
Figure 1.12: Least-squares curve fitting.
. Least Squares
Outlier
. l1 (robust )
.
. .
. .
Time
..
.
Figure 1.13: Least-squares curve fitting in the presence of outliers.

1.2 Reflections
In this section we will study the time-distance curves for seismic reflections. Reflection seismology is the
area of seismology that studies the Earths interior using reflected waves.
1.2.1 T X curves for a primary reflection
Horizontal Interface
We will compute the travel-time for a primary reflection. In the first case, we will assume a horizontal
layer over a half space.
We will assume that a P wave impinges at an angle of incidence i on the interface, the reflection will
emerge at an angle i = i (Snell Law). In Figure 3.14 we portray the geometry of the experiment. The total
travel time is given by
AD DE CE
T = + = , (1.39)
v1 v1 v1
where
CE 2 = (2zs )2 + x2 . (1.40)
The time as a function of the source receiver distance x becomes:
r
2zs 2 x
T (x) = ( ) + ( )2 . (1.41)
v1 v1
We now define the two-way zero offset travel time
2zs
T (x = 0) = T0 = , (1.42)
v1
after substituting equation (1.42) into equation (1.41) we obtain the the following final expression
x2
T (x)2 = T02 + . (1.43)
v12
The last equation can be rearranged in the following form:
x2 T (x)2
2 2 + =1 (1.44)
T0 v1 T02
1.2. REFLECTIONS 19
which is the equation of a hyperbola1 in T x.
Dipping Interface
The travel-time for the dipping interface case can be derived with the aid of Figure 3.16:
AD + DE CE
T = = . (1.45)
v1 v1
We can use the cosine theorem to write
CE 2 = (2 zs )2 + x2 4 zs x cos() , (1.46)
where
= 90o + => cos(90o + ) = sin() . (1.47)
The travel-time becomes
(2zs )2 + x2 + 4 x zs sin()
T (x)2 = (1.48)
v12
if T (x = 0) = T0 = 2zs /v1 ,
x2 4 x zs sin()
T (x)2 = T02 + + . (1.49)
v12 v12
We can re-write the latter as
T (x)2 (x + v1 T0 sin())2
2 = cos() (1.50)
T0 v12 T02
which is also a hyperbola in T x. In the above equation we have been able to write T (x) in terms of x and
zs . Where x is the source-receiver distance for a fixed source. Seismograms that share a common source (xs
is constant) and x is variable are called Common Source Gathers (CSG) or Shot Gathers.
2
1 Hyperbola: u v2
a2
b2
=1
A E
A: Source
E: Receiver
z
i i V1 C: Image of the source
B
D
V2
Figure 1.14: A primary reflection. The horizontal layer case.
x3
x2
x1
Source R1 R2 R3
Z
V1
V2
T0=2Z/V1
Time
Hyperbola
2 2 2
T = T0 + (X/V1)
Figure 1.15: Common Shot Gather(CSG).

1.2. REFLECTIONS 21
x3
x2
x1
Source R1 R2 R3
A E
zs
i i V1
B
V2
AD = DC
AB = BC
Figure 1.16: A primary reflection. The dipping interface case.

1.2.2 Relative position of the reflection and head wave in the T x plane
In Figure (1.17) we portray the T x diagram for a single horizontal layer over a half space. We can visualize
the travel-time for the following waves:
x
Direct wave: Td = v1
Head wave: Th = 2 z cos(i

v1
c)
+ vx2
q 2 2z
Reflection: Tr = T02 + xv2 , T0 = v1
1
The geological model that I used to obtain Figure (1.17) is summarized as follows:
v1 = 2000m/s
v2 = 3500m/s
z = 200m
First, we want to study the asymptotic behavior of the reflection. In other words we want to compute
d Tr
lim . (1.51)
x d x
We can start by taking derivatives of Tr
d Tr x
= 2p 2 (1.52)
dx v1 T0 /x2 + 1/v12
from where it is easy to see that
d Tr 1
lim = . (1.53)
x d x v1
In other words the asymptotic line can be written us
1
Ta = a x + b = x+b. (1.54)
v1
At we expect
lim (Tr Ta ) = 0 (1.55)

x
x
last condition is satisfied if b = 0. Therefore, the asymptotic line Ta = v1 coincides with the travel-time of
the direct wave Td .
1.2. REFLECTIONS 23
We can also compute the distance x where the times of the refraction and the reflection are equal
Td = Th
or
x 2 x2
( 2v1z cos(ic ) + v2 ) = T02 + v12
x2
= ( 2v1z )2 + v12
.
After a few manipulations we arrive to the solution x
x = 2 z tan(ic ) (1.56)
It is interesting to note that x is equal to the critical distance xc (see equation (1.22)). This is the distance
where we start to record the head wave. It is also interesting to compute the slope of the reflection at
x = x = xc
d Tr 1
|x=2 z tan(ic ) = (1.57)
dx v2
the latter shows that at x = xc the reflection and the head wave are tangent to each other.
All this information is summarized in Figure (1.17).

1.4
1.2
0.8
T [s]
0.6
Tr= (T20+x2/v21)1/2
Th = T0 cos(ic) + x/v2
0.4
0.2
Td = x/v1
0
0 500 1000 1500 2000 2500
x [m]
xc
Figure 1.17: Relative position of the direct wave (Td ), the reflected wave Tr and the head wave (Th ) in the
T x plane.
1.3 Reflections in horizontally stratified media
In this section we will analyze the time-offset response of a layered Earth model. We will assume that our
geological model is composed of a stack of n horizontal layers (Figure 1.18).
First, we will compute the travel time for a reflection generated at the n th layer. According to Snell
law
sin(i1 ) v1
=
sin(i2 ) v2
sin(i2 ) v2
=
sin(i3 ) v3
......d...
sin(in1 ) vn1
=
sin(in ) vn
now we multiply all the LHS terms and equate them to the multiplication of all the RHS terms:
sin(i1 ) sin(i2 ) sin(ik1 ) v1 v2 vk

... = ... 1 (1.58)
sin(i2 ) sin(i3 ) sin(ik ) v2 v3 vk
it is easy to see that for the k th layer the following condition is satisfied:
1.3. REFLECTIONS IN HORIZONTALLY STRATIFIED MEDIA 25
sin(i1 ) v1
= . (1.59)
sin(ik ) vk
or
sin(il ) sin(i1 )
= =p (1.60)
vk v1
we have just proved that in a horizontally stratified media the horizontal ray parameter p is constant.
In Figure (1.18) we are assuming that the emergent wave is of the same type of the incident wave (P P or
S S). In this case the total offset x is given by
n
X
x=2 hk tan(ik ) . (1.61)
k=1
Similarly, we can compute the total time as the sum of the time the wave travels in each layer
n
X hk
T =2 . (1.62)
vk cos(ik )
k=1
We can use the fact that the ray parameter is constant to simplify equations (1.61) and (1.62) and write
n
X h v p
x=2 p k k , (1.63)
k=1
1 (vk p)2
n
X hk
T =2 p . (1.64)
v
k=1 k
1 (vk p)2
The last two equation express the time of the reflection at the n th layer and the distance x in parametric
form (p is the parameter). One can specify a value of p and use (1.63)-(1.64) to build a time-distance table
for each reflector. In order to simplify the processing of seismic data, in next section we will try to find a
direct relationship between T and x (T = f (x)).
1.3.1 RMS velocity
We will attempt to find a function that related T with x. It is clear that it is impossible to find a closed
solution of the type T = f (x) by simple algebraic substitutions in equations (1.63) and (1.64). Our approach
entails finding an approximation to the function T = f (x). Lets start with the Taylor expansion of the
travel time:
x2
T (x) = T (0) + T (x)|x=0 x + T (x)|x=0 + ... (1.65)
2
Where T (0) is the time for x = 0, therefore, T (0) can be obtained as the sum of the vertical two-way
travel-time in each layer.
n
X hk
T (0) = 2 . (1.66)
vk
k=1
One can mathematically prove that T (x)|x=0 = 0. I would prefer to use a very simple physical argument
to prove this fact. Suppose that we interchange source by receiver and we repeat the seismic experiment.
The travel-time that we obtain should not change (the ray are traveling the same path regardless on the fact
that we switch receiver by source and vice versa). Therefore the following condition should be satisfied:
T (x) = T (x)
The latter indicates that all the odd power terms in the Taylor expansion should vanish
T (x)|x=0 = 0 , T (x)|x=0 = 0 . . .
If we assume that the coefficient T (x)|x=0 is small, the only coefficient that we need to compute is
T (x)|x=0 . This can be done by using the fact that (applying the chain rule)
d2 T d dT d2 T dp 2 dT d2 p dp
= ( ) = ( ) +
dx2 dx dx dp2 dx dp dx dp dx
after evaluating all the terms in the last equation at x = 0 (they can be computed from (1.64) and (1.63))
we end up
1
T (x)|x=0 = Pn . (1.67)
2 k=1 hk vk
Now, our travel-time expression is given
1 x2
T (x) = T (0) + Pn . (1.68)
2 k=1 hk vk 2
Note that
n n
X 2hk X
T (0) = = tk ,
vk
k=1 k=1
1.3. REFLECTIONS IN HORIZONTALLY STRATIFIED MEDIA 27
where tk = 2 dk /vk is the two-way transit time at zero offset in each layer. The second term in the RHS of
equation (1.68) can be written in terms of tk
1 x2
T (x) = T 0 + Pn 2 . (1.69)
k=1 tk vk 2
In order to make the last equation look like the equation that corresponds to the single layer case
(Hyperbola) we do the following trick
1 x2 2
T (x)2 = (T 0 + Pn 2 ) . (1.70)
k=1 tk vk 2
after expanding the square and neglecting fourth-order term we obtain
x2
T (x)2 = T 02 + 2
(1.71)
vrms
sP
n
k=1 tk vk2
vrms = P n . (1.72)
k=1 tk
The latter is called Root Mean Squared Velocity or RMS velocity. The seismic record is a superposition
of events that exhibit a hyperbolic moveout. The rms velocity of each layer can be identified (measured)
together with the associated T (0) and used to find the actual internal velocities.
S x R
v1 h1
i1 i1
i2 i2
v2
i2 i2
.
.
.
.
vn-1 hn-1 i n-1
hn
in in
vn
Figure 1.18: Reflections in horizontally stratified media
1.4 Seismic reflection data in offset and midpoint coordinates
In section (3.2), we derived an expression for the travel-time for a CSG for a single dipping interface.
The travel-time of a reflection can be parametrized with different variable. Figure (3.19) displays several
acquisition variables that can be used to parameterize travel-time curves.
xr Position of the receiver
xs Position of the source
The source-receiver distance is called offset,
h = xr xs Offset (1.73)
The point m in Figure (2.19) is called the mid-point, the coordinate of the midpoint is related to xr and
xg by the following expression
xr + xs
xm = Midpoint (1.74)
2
1.4. SEISMIC REFLECTION DATA IN OFFSET AND MIDPOINT COORDINATES 29
The travel-time curve can be either defined as a function of (xr , xs ) or as a function of (h, xm ):
T (h, xm ) : Time in offset-midpoint coordinates
T (xr , xs ) : Time is receiver-source coordinates
Since seismic waves are generated at different source position xs and observed at different receiver positions
xr , our data constitute a 3D volume2 (xr , xs , t) or (h, xm , t).
Our original expression for the travel-time for a dipping layer (equation (3.48)) can be written in terms
of h as follows
(2zs )2 + h2 + 4hzs sin()

T2 = (1.75)
v12
It is clear (Figure (3.19)) that zs is a function of the source position, lets try to remove the dependency
of the experiment on zs and try to write the travel-time as a function of Z the depth under the mid-point.
After analyzing Figure (3.19) we can write
h
Z = zs + sin() (1.76)
2
After substituting (1.76) into (1.75) we obtain
2Z 2 h
T2 = ( ) +( )2 (1.77)
v1 v1 /cos()
We have been able to write the travel-time in terms of h and Z, but since Z is a function of the midpoint
xm , we can always write
2Z 2 h
T (h, xm )2 = ( ) +( )2 . (1.78)
v1 v1 /cos()
Or
h
T (h, xm )2 = T02 + ( )2 . (1.79)
v1 /cos()
If our observations are acquired with an experiment where the midpoint position is kept constant (xm =
constant) we have a Common Midpoint Gather. The data cube can be organized in different gathers
according to
2 We are assuming that seismic data are acquired along a 1D line.
xr
xm
xs
Source m R
A E
Z
zs
i i V1
B
D
V2
Figure 1.19: A primary reflection. The dipping interface case.
1. CMP gather: h = variable, xm = constant. (Common Midpoint)
2. COF gather: xm = variable, h = constant. (Common Offset)
3. CS gather: xr = varialbe, xs = constant. (Common Source)
4. CR gather: xs = varialbe, xr = constant. (Common Receiver)
We will concentrate our attention into CMP gathers. These gathers are an essential part of seismic data
processing. We will use CMP gathers to construct a seismic section. The latter is an image of the
underlying geology in t xm space.
1.4. SEISMIC REFLECTION DATA IN OFFSET AND MIDPOINT COORDINATES 31
1.4.1 Multiples
Multiples are a real problem in seismic exploration. Seismic data are composed of signal plus noise. The term
noise does not only define stochastic or random noise. We can have deterministic noise or in other words
undesired waveforms. For a seismologist noise is everything that corrupts the analysis and interpretation of
seismic data.
A seismic interpreter uses primary events to construct a geological model. If multiples are not properly
eliminated we may end up with a wrong seismic image, and consequently with a wrong interpretation.
The travel-time for the direct reflection in Figure (1.20) is given by
h 2 2z
Td2 = T02 + ( ) , T0 = (1.80)
v1 v1
For the multiple reflection we have
2 h 2
Tm = (2T0 )2 + ( ) (1.81)
v1
The multiple reflection defines a hyperbola which is parametrized with the velocity v1 . The intercept time
for the multiple reflection is 2T0 . We can have higher order multiples. They will show up at intercept times
3T0 , 4T0 .... In fact, we can recognize multiples by its periodic behavior at small offsets h 0. In the dipping
layer case we have
s
h
Tp = (T0 )2 + ( )2 (1.82)
v/cos()
s
h
Tm = (2T0 )2 + ( )2 (1.83)
v/cos(2)
We note that the first order multiple simulates a layer with velocity v/cos(2) at intercept time 2T0 .
S Primary R
z V1
Multiple
2z
V2
Figure 1.20: A primary reflection and a first order multiple.
1.5 NMO correction
The travel-time for a reflection in a CMP gather is given by
r
h
T = T02 + ( )2 . (1.84)
v
where h is the offset or source-receiver distance. The NMO correction (Normal Moveout) is a correction
that is used to remove the offset dependency in equation (1.84).
r
h N MO
T = T02 + ( )2 T = T0 (1.85)
v
Before NMO (Figure (1.21) the travel-time for a primary reflection in a CMP gather is given by a
hyperbolic curve parametrized by the velocity of the layer. After NMO correction the offset dependency has
been removed, this is like simulating a seismic experiment where each source-receiver pair shares the same
midpoint location and same offset, h = 0.
After NMO correction all traces should look alike (this is not completely true, since Im neglecting noise,
amplitude factors, and multiples). One can stack all the traces associated to a given midpoint position to
obtain a Stacked Trace. Mathematically, we can think in a procedure of the form:
1.5. NMO CORRECTION 33
Before NMO
s1 s2 s3 r1 r2 r3
h
T0
T(h)
CMP Fold = 3 (3 Traces per CMP)

t
After NMO
s1 r1
(Source-Receiver in the same postion : h=0)
s2 r2
s3 r3 h
T(h=0)=T0
Figure 1.21: NMO correction attempts to simulate a seismic experiment where all the source-receiver pairs
at a given mid-point position have offset h = 0.
Data Before NMO
DAT A(xm , h, t) Seismic volume before NMO
A fix position xm corresponds to a CMP gather before NMO.
Data After NMO
DAT A(xm , h = 0, t) Seismic volume after NMO
A fixed postion xm corresponds to a NMO-corrected CMP gather.
Data After NMO and Stacking
DAT A(xm , t) Seismic volume after NMO
Notice that now a fixed portion xm is a single seismic trace. It is clear that after applying NMO+Stacking
we have remove one dimension to our data volume
(xm , h, t) (xm , t) .
A collection of Stacked Traces each one having a different mid-point position is what we call a Seismic
Section. The seismic section is an image in time of the earths interior. Each trace in a seismic section
corresponds to the trace one should have obtained after running an experiment where the midpoint position
is variable whereas the source-receiver distance (offset) is kept constant and equal to zero (h = 0).
To summarize:
1. The NMO correction removes the dynamic term h2 /v 2 in the hyperbolic travel-time curve
2. NMO correction followed by STACKING maps the data cube h, xm , t into xm , t
3. NMO correction followed by STACKING enhances the signal-to-noise-ratio (Figures (1.5),(1.23), and
(1.5)).
4. NMO correction followed by STACKING attenuates multiple reflections (we will see this in more detail
when dealing with multiple attenuation
1.5. NMO CORRECTION 35
CMP gather sdev=0.1 CMP after NMO CMP after NMO and stacking
0 0 0
0.1 0.1 0.1
0.2 0.2 0.2

time (sec)
0.3 0.3 0.3
0.4 0.4 0.4
0.5 0.5 0.5
200 0 200 200 0 200 0 2 4 6

offset (m) offset (m)
Figure 1.22: NMO and stacking.
CMP gather sdev=0.5 CMP after NMO CMP after NMO and stacking
0 0 0
0.1 0.1 0.1
0.2 0.2 0.2

time (sec)
0.3 0.3 0.3
0.4 0.4 0.4
0.5 0.5 0.5
200 0 200 200 0 200 0 2 4 6


CMP gather sdev=0.2 & Coherent noise CMP after NMO CMP after NMO and stacking
0 0 0
0.1 0.1 0.1
0.2 0.2 0.2

time (sec)
0.3 0.3 0.3
0.4 0.4 0.4
0.5 0.5 0.5
200 0 200 200 0 200 0 2 4 6


Chapter 2
Reflection seismology: NMO, VA and

Stacking
In the previous chapter we have analyzed the basic travel-time curves for refraction and reflection data. We
have introduced the concept of CMP gathers and examined, briefly, the NMO correction. In this chapter we
will further explore these concepts and discuss several procedures to construct zero offset sections1
2.1 Stacking charts
In this section I would like to mention a few words about seismic acquisition. Conventional 2D seismic
lines are acquired by deploying a large number of receivers along a line and moving the source of energy
(explosives or vibrosies). In general only a set of receivers are active for a given source. In Figure (2.1) I
portray a simple acquisition geometry that corresponds to N S = 20 sources (see vertical axis indicating the
source number or field record number) and N R = 24 receivers per shot. The axis x indicates the position
of the source or receiver for each field record. The geometry of the experiment can be summarized with the
following variables:
Distance between shot points: s = 24 m
Distance between receivers: r = 25 m
Near offset (Minimal distance source-receiver): hmin = 100
Distance between consecutive CMP positions xm = s/2 = 12.5 m
In this example the first shot point is located at xo = 1000 m. We can express the source position as follows
1 Also called Stacked Sections; these are a time seismic image of the subsurface.
37
38 CHAPTER 2. REFLECTION SEISMOLOGY: NMO, VA AND STACKING
xs (i) = xo + (i 1) s , i = 1, . . . , N S .
Similarly, the position of the receiver j when source i is used is given by
xr (i, j) = xs (i) + hmin + (j 1) r , j = 1, . . . , N R .
In Figure (2.1) we display the source-receiver position for each seismic trace. The rotated axes correspond
to traces with the same midpoint position (CMP gather) and traces with equal offet (CO gather). During
acquisition, seismic traces are stored in field records. In other words, the data are organized in a binary file2
with the following structure:
Header/Trace Shot 1 Receiver 1

.
.
.
.
.
.
Header is a 240 byte partition where all the relevant information needed to caracterize the Trace is
recorded (i.e., source, receiver position, offset, CMP number, time sampling interval, etc).
A procedure called sorting permits one to re-orginize the seismic volume in CMP gathers. After CMP
sorting the ouput will look like
Header/Trace CMP 1 Offset 1

.
.
.
2 SEGY file: this is a special file format to handle seismic data.
2.2. VELOCITY ANALYSIS 39
Stacking chart
2
xm =1437.5m, fold =12
4
8
Field Record (shot number)
10
12
14
16
xm =1775m, fold =2
18 s =25m, r =25 m
Near offset =100 m
20 Max Fold =12
xm =12.5 m
22
1000 1200 1400 1600 1800 2000 2200 2400

x position (m)
Figure 2.1: A stacking chart for a experiment that consists of 20 sources and 24 receivers. The receivers
participating in the formation of the CMP gathers with midpoint position xm = 1437.5m and xm 1775.5m
are also indicated. Note that the maximum fold in 12 (number of traces per CMP).

.
.
.
where CMP 1 indicates the CMP number at the first midpoint xm position.
2.2 Velocity analysis
We have already seen that in a CMP gather the moveout curve of a reflection can be approximated by a
hyperbola:
SourceReceiver map
1600
CO Gather
1500
CMP Gather
1400
1300
xs [m]
1200
1100
1000
900
1100 1200 1300 1400 1500 1600 1700 1800 1900 2000 2100
x [m]
r
Figure 2.2: Source-receiver position for the experiment displayed in the previous Figure. Each dot indicates
a seismic trace. The rotated axes indicate the traces participating in a CO gather and a CMP gather,
respectively.
q
T = T02 + h2 /v 2 (2.1)
3
where h is the source-receiver distance (offset), v is NMO velocity of the reflection, and T0 is the two-way
zero offset traveltime. The term Velocity Analysis refers to a numerical algorithm that permits retrieval
of the pairs T0 , v that parametrized the seismic reflections in a CMP gather.
Velocity analysis is performed in consecutive temporal windows by evaluating a coherence measure along
a tentative moveout curve. The analysis is carried out for a set of velocities; the velocity corresponding
to the peak of the coherence measure is interpreted as the velocity that best flattens the seismic event (the
velocity you will use for NMO corrections).
If C(t0 , v) denotes the coherence at a window centered at t = t0 , an estimate of the velocity v is obtained
at the peak value of C(t0 , v). A standard coherence measure is the semblance. To derive the semblance,
we first define the data window of length 2 M + 1 centered about a moveout curve with intercept time, ti ,
and velocity, vj

xti Mt,1 xti MT,2 . . . xti MT,N
M(ti , vj ) = ... ... ... ... . (2.2)
xti +Mt,1 xti +MT,2 . . . xti +MT,N
where N indicates the number of traces and T the time sampling interval of the seismogram.
The semblance is given by the following expression
P P 2
t( Pk xt k )
S(ti , vj ) = P 2 . (2.3)
k t xt k
A pseudo-code for velocity analysis should look like
FOR all ti and vj
Construct the matrix M(ti , vj )
Compute coherence measure ,i.e., S(ti , vj )
ENDDO
It is clear that when the one of the tentative velocities matches the velocity of the reflection, the seismic
event in the window of analysis will be horizontal, and therefore, the semblance will exhibit a peak.
A CMP gather and its semblance panel are portrayed in Figure (2.3). This CMP corresponds to a land
survey in the Western Canadian Sedimentary Basin (WCSB); in this example it is quite hard to identify
multiple reflections.
3 They should be close to rms velocities
Offset (m) vnmo (m/s)

-2000 -1000 0 1000 2000 2000 3000 4000
0 0
0.2 0.2
0.4 0.4
0.6 0.6
0.8 0.8
1.0 1.0
t (s)
t (s)
1.2 1.2
1.4 1.4
1.6 1.6
1.8 1.8
2.0 2.0
CDP 750 CDP 750 (Semblance)
Figure 2.3: A CMP gather and its associated semblance panel.

2.2.1 Velocity Scans
Velocity scans are computed by displaying the CMP gather after applying a constant velocity NMO correc-
tion4 . By displaying consecutive NMO corrected scans we can visualize the velocity and the time at which
event becomes horizontal. After displaying consecutive scans we can easily pick the T0 , v pairs. An algorithm
to compute velocity scans looks like
FOR all vj = vmin + v
Apply constant NMO correction to the CMP gather with v = vj
Display Scan j
ENDDO
In Figure (2.4) I portray the velocity scan for the CMP in Figure (2.3). The first panel was computed using
a velocity v = 1500m/sec, the second with v = 1750m/sec and so on. You can try use Figures (2.3) and
(2.4) to pick the NMO velocities.
4 Constant velocity NMO correction, means that the same velocity is used for all t
CDP 750 (vmin=1500m/s vmax=4000m/s dv=250m/s)
300
250
200
NMO Scan
150
100
50
0 0.5 1.0 1.5 2.0

Time (s)
Figure 2.4: Velocity scan for the CMP portrayed in Figure (2.3), the minimal NMO velocity is 1500 m/s,
the maximum velocity is 4000 m/s. We have adopted a velocity interval of 250 m/s.
2.2.2 Constant velocity stacks
Constant velocity stacks are obtained by NMO correcting a set of adjacent CMP gathers and stacking each
CMP along offset. This procedure produces a small stacked section for each NMO velocity. We can use
constant velocity stacks to retrieve the velocities that optimize the stack for the selected CMP gathers.
The flow chart looks like:
FOR cdp=1,N (pick N adjacent CMPs)
FOR all vj = vmin + v
Apply constant NMO correction to the CMP gather with v = vj
Stack the NMO corrected CMP along offset
ENDDO
Display Constant Velocity Stack j
ENDDO
In Figure (2.5) I portray an ensemble of Constant Velocity Stacks computed using CMPs 640 to 660.
2.2.3 The seismic section
Velocity analysis, velocity scans and constant velocity stacks are procedures used to estimate NMO velocities
at given cmp locations. In general, this is reapeated for a set of CMP positions to produce an NMO velocity
model. Lets say that NMO velocities are needed every 50 CMPs. The NMO velocities for the rest of the
CMPs are estimated via interpolation.
Once we have picked the NMO velocities for a set of mid-point positions, the remaining process entails
the application of NMO corrections to all the CMP gathers and stacking. Each CMP will generate a stacked
trace, the seismic section is composed of all these stacked traces plotted versus mid-point (or CMP/CDP
number). It is important to stress that the stacked section (for simple geological models) resembles the
section one should have obtained after running a zero-offset experiment. This is an experiment where source
and receiver occupy the same field position.
These are the velocities that I picked5 .
cdp=545,645,745,845
tnmo=0.5954,0.8043,0.9087,0.9366,1.058,1.18,1.431,1.581,1.856,2.005
5 The first line is the CDP number where velocity analysis was carries out
File: cdps.su Constant-Velocity Stack
3000
2800
2600
Velocity (m/s)
2400
2200
2000
1800
1600
0 1 2
Time (s)
Figure 2.5: Constant velocity stacks for CMP 640 to 660.

vnmo=2674,2964,3112,3135,3175,3249,3477,3517,3477,3449
tnmo=0.6058,0.8878,0.9958,1.152,1.737,1.852,1.988
vnmo=2691,3158,3232,3369,3608,3716,3722
tnmo=0.3934,0.9401,1.1,1.191,1.351,1.619,2.009
vnmo=2657,3243,3386,3426,3563,3534,3722
tnmo=0.5188,1.062,1.163,1.936,2.002,2.103
vnmo=2919,3375,3414,3563,3580,3631
Final Seismic Section

300
250
200
cmp
150
100
50
0 1 2
Time (s)
Figure 2.6: Final seismic section obtained after NMO correction and stacking all the CMP gathers
2.3. STACKING AND ITS FREQUENCY RESPONSE 49
2.3 Stacking and its Frequency Response
We will analyze the frequency response of the stacking procedure when applied to a CMP gather after the
NMO correction.
Consider the t h diagram for a primary reflection and for a multiple reflection. We will assume that
the intercept t0 is the same for both reflections. The velocity of the primary reflection is denoted by vp and
the velocity of the multiple by vm . The t h relations are
2 h2
Tm = T02 + 2
(2.4)
vm
h2
Tp2 = T02 + (2.5)
vp2
Now suppose that the NMO correction was applied, and that the primary reflection has been properly
flattened. The residual moveout or time diference between the primary and the multiple is given by:
M = Tm Tp (2.6)
The traveltime for the multiple can be written as follows:
2 h2
Tm T02 = (Tm + T0 )(Tm T0 ) = 2
. (2.7)
vm
If we assume that Tm T0 , we can write
h2
2(T0 )(Tm T0 ) 2
(2.8)
vm
From where it is easy to see that
1 h2
(Tm T0 ) 2
(2.9)
2T0 vm
A similar trick is used to obtain an expression for Tp ,
1 h2
(Tp T0 ) . (2.10)
2T0 vp2
Now that we have been able to eliminate the square roots, we can find the following expression for the
residual moveout
h2 1 1
M= ( 2 2 ). (2.11)
2T0 vm vp
This is the expression for a parabola in h t. Consider the residual moveout at far offset (h = hmax ), and
the residual moveout at any generic offset h,
h2max 1 1
Mmax = ( 2 2 ). (2.12)
2T0 vm vp
we can write the residual moveout at the trace i-th with offset hi as follows
h2i
Mi = Mmax . (2.13)
h2max
When we design a seismic experiment we need to consider various acquisition parameters. For instance, we
need to define the minimum and maximum offset (near and far offset traces) and the fold (number of traces
per CMP). We will examine the relationship between these variables and how one can select them in order
to optimally attenuate multiples.
Let us assume that the differences in offset are constant:
h4 h3 = h3 h2 = h2 h1
In this case our fold is N = 3 (our CMP is composed of 3 seismograms). We will need the following variables
(see figure)
L = hmax hmin
hmin = KL
where K is a constant to determine. The increment in offset can be written as
L
h4 h3 = h3 h2 = h2 h1 = ,
N 1
the offset of the i-th trace is given by
L
hi = K L + (i 1) , hmax = K L + L = (K + 1)L . (2.14)
N 1
Now, we recall the expression for the residual moveout

h2i K(N 1) + (i 1) 2
Mi = Mmax = Mmax ( ) , i = 1, 2, ...N (2.15)
h2max (K + 1)(N 1)
The residual moveout Mi is the time difference between the primary and the multiple at receiver i after
NMO correction. If we stack our CMP gather after NMO correction, it is clear that the primary will be
stacked in phase or in other words we will make the primary stronger while the multiple will be stacked out
of phase and we will make it weaker. Can we measure to what extent the multiple has been attenuated after
stacking? In order to answer this question, first, we will consider that the primary and the multiple can be
modeled using a simple complex exponential of frequency f and amplitude.
Primary after NMO (the delay between traces is 0 everywhere).
xp (hi ) = Aei2f 0 = A (2.16)
Multiple after NMO (the delay is now Mi ).
K(N 1)+(i1) 2
xm (hi ) = Aei2f Mi = Aei2f Mmax ( (K+1)(N 1)
)
(2.17)
The frequency response of the stacking procedure can be evaluated via the following expression:
Enery of the M ultiple af ter N M O Em

R(f ) = = (2.18)
Enery of the P rimary af ter N M O Ep
where
N
X N
X
Ep = |xp (hi )|2 = A2 = N A2 (2.19)
k=1 k=1
N
X
Em = |xm (hi )|2 (2.20)
k=1
The response function becomes:
N
Em 1 X i2f Mmax ( K(N 1)+(i1) 2
) 2
R(f ) = = |e (K+1)(N 1) | (2.21)
Ep N
k=1
In general we need some prior knowledge of the velocities of the reflections and multiples in order to find an
upper bound for M . In other words we need to know the residual moveout at far offset. Once that we define
hmax we compute Mmax and we compute the response function for given values of K and N . We iterate
h3
h2
h1
s3 s2 s1 r1 r2 r3
Reflector that generates the multiple
Reflector that generates the primary
s: source r: receiver h: offset N=3 (fold)

Interference of primary and multiple
r1 r2 r3
T0 Primary after NMO
M:Residual Moveout
Multiple after NMO
Figure 2.7: A primary event and a multiple after NMO correction. The velocity of the primary was used to
perform the NMO correction. The residual moveout is designated by the variable M .
until we find a curve that rejects the energy of the multiple in the seismic band (5 100Hz). Once we have
K and N it is easy to compute the minimum offset and the distance between receivers in the field. NMO
followed by stacking acts as a filter. This is an interesting type of filter since it depends on the parameters
used to acquire the data (K,N ,hmax ).
R(f,N,K): Frequency response = energy of the multiple/energy of the primary
dc value (0Hz component of the signal) is not attenuated.
1
R(f,N,K)
fmin fmax f
0
Band of attenuation
Figure 2.8: Response as function of frequency. The response curve measures the ratio of energy of the
multiple events to the energy of the primary after NMO correction. Acquisition parameters are computed
in order to obtain maximum attenuation in the seismic frequency band.
Chapter 3
Modeling Normal Incidence

Seismograms
In this chapter we will study the problem of computing reflection and transmition coefficients for a layered
media when plane waves propagate in the earth with angle of incidence i = 0 (Normal incidence).
Assume an interface separating two media with P velocities v1 and v2 and densities 1 and 2 , respectively.
3.1 Normal incidence
Consider a plane wave impinging at an angle of propagation i = 0 with respect to the normal (see Figure
(3.1) ). In this case we have three waves:
Incident wave: in medium 1
Reflected wave: in medium 1
Transmitted wave: in medium 2
Let us assume that the amount of incident wave is equal to 1, the amount of reflected wave is given by
r, and the amount of transmitted wave is denoted by t. At the boundary the following condition should be
satisfied (continuity of displacements)
1+r =t
This equation has two unknowns, to compute the r and t we need an extra equation. We will consider
conservation of energy. In the acoustic (vertical incidence case) conservation of energy leads to the following
equation:
55
56 CHAPTER 3. MODELING NORMAL INCIDENCE SEISMOGRAMS
I1 12 = I1 r2 + I2 t2 .
The quantities I1 and I2 are called acoustic impedances
I1 = 1 v1 , I2 = 2 v2
where 1 and 2 are the densities of the material above and below the interface and v1 and v2 the P-
velocities, respectively. After combining the equations of continuity of displacement and conservation of
energy we obtain the following expressions
I1 I2
r = Reflection coefficient (3.1)
I2 + I1
2I1
t = Transmition coefficient (3.2)
I2 + I1
The above analysis is valid for an incident plane wave propagating downwards. Lets consider the case of an
incident wave propagating upwards (Figure (3.2) ).
Incident wave: in medium 2
Reflected wave: in medium 2
Transmitted wave: in medium 1
In this case the reflection and transmition coefficients are given by
I2 I1
r = , (3.3)
I2 + I1
2I2
t = . (3.4)
I2 + I1
From the above equations it is clear that
r = r (3.5)
3.1. NORMAL INCIDENCE 57
1 r
I1
I2
r: reflection coefficient
t: transmition coefficient
1+r=t
Figure 3.1: P-wave normal incidence. The incident wave is propagating downwards.
I1
I2
1 r
r: reflection coefficient
t: transmition coefficient
1+r=t r = -r
Figure 3.2: P-wave normal incidence. The incident wave is propagating upwards.
3.1.1 Impulse response
Lets assume that we run a zero offset experiment in a stratified earth composed of 4 layers plus a half-space
of impedances given by I1 , I2 , I3 , I4 , I5 . (Figure (3.3) ). At t = 0 a delta-like source emits energy into Earth.
The energy is transmitted and reflected from the layers. If we do not consider multiples, our sismogram will
be composed of 4 arrivals (4 reflections).
To simplify the problem I will show how to compute the amplitude of the wave recorded at the surface
of the earth generated (reflected) at the interface 4. First we have to compute the amount of amplitude
transmitted to each one of the layers until reaching layer 4. This is given by the product of the transmi-
tion coefficients of each layer. In Figure (3.3) the transmition coefficients t are replaced by the equivalent
expression (1 + r).
The amplitude of the wave reaching layer 4 is given by the following expression:
1 t1 t2 t3 = (1 + r1 )(1 + r2 )(1 + r3) ,
when the wave is reflected at interface 4 the total amplitude at that point (last expression) needs to be
multiplied by the reflection coefficient of interface 4, in other words
1 t1 t2 t3 r4 = (1 + r1 )(1 + r2 )(1 + r3)r4
Note that now the wave (reflected wave) is propagating upwards, therefore, the transmition coefficients are
given by terms of the form
1 + r = 1 r .
The final amplitude after propagating the wave to the surface of the Earth (this is what the receiver is
measuring!) is given by
(1 + r1 )(1 + r2 )(1 + r3 ) r4 (1 r1 )(1 r2 )(1 r3 ) .

| {z } |{z} | {z }
Transmition Reflection Transmition
The final expression for the amplitude of the wave reflected in the interface 4 can be written down as follows
(1 r12 )(1 r22 )(1 r32 )r4 .
It is clear that reflections occur at all the layers:

z=0
1 (1+r1)(1+r2)(1+r3)r4(1-r3)(1-r2)(1-r1)
I1
Interface 1
z1
1+r1 (1+r1)(1+r2)(1+r3)r4(1-r3)(1-r2)
I2
z2
(1+r1)(1+r2) (1+r1)(1+r2)(1+r3)r4(1-r3)
I3
z3
(1+r1)(1+r2)(1+r3) (1+r1)(1+r2)(1+r3) r4
I4
Reflection at interface 4
z4
Interface 4
I5
Ii = Acoustic impedance of the layer i
1+ri = Transmition coef.
1-ri = Transmition coef.
Figure 3.3: Amplitude of a wave plane wave propagating in a layered medium. Analysis of the wave reflected
in the interface 4.
Amplitude of the reflection generated at interface 1
A1 = r1
A2 = (1 r12 )r2
A3 = (1 r12 )(1 r22 )r3
A4 = (1 r12 )(1 r22 )(1 r32 )r4
We can write a general expression for the amplitude of a reflection generated at the k-th interface:
A1 = r1
k1
Y
Ak = (1 ri2 ) rk , k = 2, 3, 4, . . .
i=1
How do we interpret these results?. If we assume that the earth is excited with a delta function, and neglecting
the presence of multiples, our zero-offset seismogram will be a collection of delta functions (spikes) at arrival
times given by the two-way travel time formula. The strength of each arrival will be proportional to the
amplitude Ak .
However, in real exploration seismology, it is impossible to have a source that resembles a delta function.
The source signature is called a wavelet. This is a finite length time function that we will denote as w(t). In
this case, the seismogram is represented by a superposition of wavelets arriving at different times and with
amplitude proportional to Ak .
In our model with 4 interfaces (Figure (3.3) ) we will have 4 arrivals of amplitude A1 , A2 , A3 and A4 .
The seismogram can be expressed as follows
s(t) = A1 w(t 1 ) + A2 w(t 2 ) + A3 w(t 3 ) + A4 w(t 4 ) (3.6)
1
where i , i = 1, 4 are the arrival times of each reflection In GEOPH-426 we will see that the last equation
can be expressed as a convolution sum. This is a pivotal concept in linear system theory.
1 Notice that w(t ) is w(t) after being delayed seconds.
10000
I (g./cm2 m/s)
8000
6000
4000
2000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Reflection coefficient
0.5
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Time (s)
0.5
Seismogram
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Figure 3.4: Acoustic impedance versus two-way travel time for the example in Table (3.1) (top). The
reflectivity sequence (center). The synthetic seismogram (bottom).
3.1.2 Example
Lets put some numbers to the configuration in Figure (3.3).
Layer z(m) v (km/s) (g/cm3 )

1 200 1.5 1.0 Water
2 500 2.0 2.4 limestone
3 630 3.0 2.4 sandstone
4 700 2.1 2.3 limestone
Half-space 1300 4.3 2.1 Hard deeper layer
Table 3.1:
Figure (3.4) portrayes the acoustic impedance (I = v ); the reflectivity series and the synthetic
seismogram obtained by convolving a synthetic source function with the reflectivity series. In this example,
we have not considered transmittion effects and multiple reflections.
Chapter 4
Deconvolution
The classical model for the seismic trace states that we can represent the seismic trace as a convolution
integral
Z
s(t) = w( t) r( ) d (4.1)
where
s(t): seismic trace
w(t): source wavelet
r(t): reflectivity
the varialbe t indicates two-way-traveltime (TWT). Information about the subsurface is embedded in the
reflectivity series. The process of recovering r(t) from s(t) is called deconvolution.
In seismic deconvolution we are trying to recover r(t), but in general this is not possible. The best one
can do is to recover a good approximation to r(t) that we will denote r(t). We will assume that the wavelet
was properly estimated or measured from the data. This is usually not true; in general all we can estimate
is an approximation to the wavelet based on an assumption about the wavelet phase spectrum.
The deconvolution process can be carried out in the time domain or in the frequency domain. We will
deal with deconvolution in the frequency domain. The Fourier transform of a signal x(t) is given by
Z
X() = x(t) eit dt . (4.2)

63
64 CHAPTER 4. DECONVOLUTION
We will indicate that X() and x(t) are related via the Fourier Transform via the following notation:
x(t) X()
The Fourier convolution theorem states that convolution in the time domain is equivalent to multiplication
in the frequency domain. Therefore, the convolution integral that expresses the seismic trace in equation
(4.1) can be transformed into the multiplication of two complex functions:
s(t) = r(t) w(t) S() = R() . W ()
where

s(t) S()

w(t) W ()

R(t) R()
It is clear that convolution is now a simple multiplication of two functions in the frequency domain.
Therefore, the Fourier transform of the reflectivity series can be estimated by a simple spectral division:
S()
S() = R() . W () R() = (4.3)
W ()
The problem with the above spectral division is that the amplitude spectrum of the wavelet can be zero (or
very small) at high frequencies and therefore the deconvolution process could amplify the noise in the data.
In general, the spectral division is regularized1
S() .W ()
R() = (4.4)
|W ()|2 +
where is a small positive number used to guarantee the stability of the division. In the previous expression,
|W ()|2 = W () . W () is the squared amplitude spectrum of the wavelet; W () is the conjugate of W ().
It is clear that now R() is an estimate of the reflectivity in the frequency domain. In order to compute
the reflectivity in the time domain, we need to use the Inverse Fourier Transform.
1 Regularization methods are methods used to obtain a stable solution of an inverse problem.
65
(a) (b)
2 0.2
1 0.1
0 0
1 0.1
2 0.2
3 0.3
0 0.05 0.1 0.15 0 0.1 0.2 0.3 0.4 0.5
t [s] t [s]
(c) (d)
1 8
6 mu=100
0.5
4 mu=1
0
2 mu=0.1
0.5
0 mu=0.00001
1 2
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
t [s] t[s]
Figure 4.1: (a) Wavelet. (b) Reflectivity. (c) Noise-free seismogram. (d) Deconvolution with different values
of (pre-whitening parameter)
Z
1
r(t) = R() eit d . (4.5)
2
The algorithm to deconvolve the data will be numerically implemented using the Discrete Fourier
Transform (DFT). The discrite Fourier transform is used to compute the Fourier transform of discrete
time series (seismograms that after digitization are treated as discrete time series.)
In Figures 4.1 and 4.2 I provide deconvolution examples for clean and noisy data.
66 CHAPTER 4. DECONVOLUTION
(a) (b)
2 0.2
1 0.1
0 0
1 0.1
2 0.2
3 0.3
0 0.05 0.1 0.15 0 0.1 0.2 0.3 0.4 0.5
t [s] t [s]
(c) (d)
0.6 8
0.4 6 mu=100
0.2
4 mu=1
0
2 mu=0.1
0.2
0 mu=0.00001
0.4
0.6 2
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
t [s] t[s]
Figure 4.2: (a) Wavelet. (b) Reflectivity. (c) Seismogram contaminated with white noise (additive). (d)
Deconvolution with different values of (pre-whitening parameter)
Chapter 5
The tomographic problem and

introduction to linear inverse theory
So far we have been dealing with seismic reflection seismology and with the problem of constructing a seismic
section. This is not over, we still need to see how to convert a time section into a depth section. But, first
I would like to spend a couple of lectures discussing inverse problems in seismology. I will start with the
tomographic problem and later, deal with the inversion of velocity stacks and multiple suppression.
5.1 Inverse Theory
Inverse theory provides a mathematical framework to solve ill-posed problems. These are mathematical
problems where the solution does not exist or the solution is unstable.
In this section we will briefly study the tomographic inverse problem. Bear in mind that there are other
geophysical problems that can be approached using the inverse problem formalism, i.e., inversion of gravity
and magnetic anomalies, free oscillation of the Earth, etc.
We can talk for hours about inversion, in simple terms suppose we have some data d, that is related to
some earth model m via some operator F ,
F m = d,
1
in inversion we look for a stable solution to the last problem. Suppose that the data d is perturbed an
amount d, we want to find a solution
m = F 1 d
1 noise, systematic errors, etc!
67
68CHAPTER 5. THE TOMOGRAPHIC PROBLEM AND INTRODUCTION TO LINEAR INVERSE THEORY
such that if d d + d, then m m + m where m is small. In other words, we want to avoid

amplification of noise or errors in the solution. In fact, we often want the solution m to exhibit certain
features. For instance, if we are inverting for velocity, we want a positive solution.
5.2 Tomography
The goal in tomographic inversion is to determine a velocity field from a multi-source multi-receiver experi-
ment. Figure (5.1) provides an example of a tomographic experiment. The geological model is illuminated
with seismic rays. The geological structure is approximated by a mathematical model composed of nx nz
cells. The velocity within each cell is constant. A perturbation traveling from the source i to the receiver j
will take a time t to reach the receiver. Let us denote the time to go from the source i to the receiver j ,
tsi ,rj . The travel-time will depend on the velocity structure of the medium. The goal here is to determine
the velocity of the geological formation from travel-times observed by means of the tomography experiment.
If we assume that the velocity variation within the model is small, we can use the straight ray approx-
imation. In this case, the travel-times are given by
ncells
X
Tsi ,rj = l(k)i,j sk
i
where sk is the slowness in each cell and l(k)i,j are the segments on which the ray travels on in each cell
(Figure (5.1). If we consider several rays we can write the tomographic problem in matrix form:
As = t (5.1)
where t is a vector containing the travel-times and s vector containing the slowness of the cells. The matrix
A is called the tomographic matrix. This matrix is defined by the experiment. In other words, to obtain A
we need to know the position of the receivers, the position of the cells and the size of the tomographic grid.
In general the travel-times are contaminated with noise (i.e., picking errors). In this case we can write
As = t + n = tn (5.2)
Where tn are noisy travel-times (this is what you measure). In general we can assume that the errors are
Gaussian and uncorrelated. In this case the optimum solution for the tomographic problem is given by the
standard least-squares inverse. Here Im assuming that we have more data than unknowns which is not
always true. The least squares solution is given by
s = (AT A)1 AT tn (5.3)

5.2. TOMOGRAPHY 69
R1

S1 * R2
S2 *
V2 R3
S3 *
V1
: Receiver *: Source
V1: backgorund velocity V2: anomaly
V2=V1+ V
Figure 5.1: Tomographic experiment.
The story does not finish here. The matrix AT A is an ill-conditioned matrix (some of its rows are linearly
or quasi-linearly dependent). This is why we need to regularize the problem or in other words we need to
make the problem well-posed.
A simple trick is to add a perturbation to the main diagonal of the matrix (AT A). If I is the identity
matrix, and is a small scalar ( > 0) the regularized solution is given by:
s = (AT A + I)1 AT tn (5.4)
The parameter is called the regularization parameter, the trade-off parameter or the damping parameter.
In filter theory this parameter is often called the pre-whitening parameter (this is something coming from
the deconvolution jargon.).
The effect of the least-squares inversion is seen in Figures (5.4b) and and (5.4b). You can see that the
technique has not been able to recover the true model (5.3b). This is because this is an inverse problem, and
therefore, the solution is non-unique. In fact, in Figure (5.4b) we show the solution of the same problem,
but now using smoothing constraints on the vertical an horizontal derivatives of the slowness field. Both
solution are honoring the data (they reproduce the observations) and therefore they are both valid solutions.
Which one should you pick? It all depends on prior information, or in other words, on external information
(i.e., velocities extracted from bore-hole data, geological information, etc).
Lexicographic arragement of cells.

1
NX cells 2
1 2 3 4 5 6 7 R 3
NZ cells
.
.
S .
.
34 35 36 .
31
Number of cells = NX . NZ
32
cell 1 --> v1
cell 2 --> v2 33
.
. 34
.
cell 32 --> v36
35
36
Figure 5.2: Tomographic experiment. Lexicographic arrangement of cells.
Figure 5.3: Tomographic experiment. a). Distribution of rays. b) Model that we would like to retrieve from
travel-time observations (From Li and Ulrych, SEG 1997)
.
5.2. TOMOGRAPHY 71
Figure 5.4: Tomographic experiment. a) least-squares solution with damping. b) In this case the norm of
the vertical and horizontal derivatives of the model were minimized.
Chapter 6
Migration
In this chapter I will provide an overview of the pivotal process called migration. The migration process is
used to locate seismic reflectors in the right position, it is also a mean to increase lateral resolution.
We will first introduce the concept of exploding reflectors and deal with the post-stack migration
case. The seismic section computed after NMO correction and stacking is artificially simulating a zero-offset
experiment (source and receiver are in the same location). I will used the term post-stack migration and
migration of zero-offset data as equivalent concepts.
6.1 Exploding reflectors
This is a powerful analogy that can be used to simplify various wave propagation problems. Figure (6.1)
shows a field experiment where the data have been acquired via repeated zero-offset experiments. In fact,
this is obtained after processing CMP gathers. 1 .
In Figure (6.1) we show a fictitious experiment where the sources are located on top of the reflector. This
is the exploding reflector case. Both experiments are equivalent in the sense that they will produce the same
wavefield. If the travel time in the first experiment is divided by two, the two wave-fields obtained by these
two experiments are equal. We can instead assume that the velocity of the medium in the real experiment
is half its true value (v = v/2).
The exploding reflector analogy permits us to migrate post-stack data (D(xm , h = 0, t)). The exploding
reflector analogy assumes that the real experiment can be replaced by a fictitious one where only up-going
waves generated by the hypothetical explosion of the reflector are considered.
We can now define migration as the operation that maps the data recorded at the surface of the Earth
(z = 0) into an image of the subsurface. If the wavefield recorded at z = 0 is denoted by p(t, x, z = 0) the
migration process is used to compute the image as follows:
1 We have already studied that CMP gathers after NMO and stack simulate a zero offset experiment
73
74 CHAPTER 6. MIGRATION
Xm
S-R
Real Experiment Downgoing wave (Source to reflector)
Upgoing wave ( Reflector to receiver)
t1
Figure 6.1: A zero offset experiment.

6.1. EXPLODING REFLECTORS 75
Xm
R R R
S Exploding reflector
S S
Hypothetical Experiment Upgoing wave ( reflector to receiver)
t1 /2
Figure 6.2: The exploding reflector analogy.

I(x, z) = p(t = 0, x, z)
Where I(x, z) is an estimate of the reflector strength (the reflectivity of the layer). The exploding reflector
analogy implies that the reflector explodes at t = 0. Therefore, the wavefield at t = 0 is nothing else than
the image of our fictitious sources located on the reflector. The core of the migration process is an algorithm
that permits to propagate our wavefield recorded at z = 0 (data) into the subsurface.
In general, one can say that a migration algorithm is composed of two components:
1. A procedure to extrapolate wavefields in depth.
2. An imaging condition.
The imaging condition is described above: reflectors explode at t = 0. Next, we will describe a simple
procedure to extrapolate wavefields in depth.
6.1.1 The wave equation in (kx , ): introduction to one-way propagators
We will start our analysis by considering the acoustic wave equation in a 2D earth:
1 2
pxx + pzz = ptt , (6.1)
v2
The wavefield p(t, x, z) can be transformed into the Fourier space as follows:
Z
1
P (, x, z) = p(t, x, z)eit dt (6.2)
2
So far we have only applied the Fourier transform to the temporal variable t. The wave equation in the
3
(, x, z) domain is given by .
2
Pxx + Pzz = P, P = P (, x, z) . (6.3)
v2
Now we apply again the Fourier transform but with respect to the spatial variable x.
Z
1
P (, kx , z) = P (, x, z)eikx x dx (6.4)
2
The wave equation in (, kx , z) is given by

2p 2 P
xx = x2
3I dn f (x)
have used the following property: dxn
(i)n F ()
2
kx2 P + Pzz = P, P = P (, kx , z) . (6.5)
v2
We can rearrange the above equation as follows:
2
Pzz = ( kx2 )P (6.6)
v2
The dispersion equation of the scalar wave equation is given by
2
kx2 + kz2 = (6.7)
v2
where kz is the vertical wavenumber. In equation (6.6) we can recognize the vertical wavenumber kz2 =
2
v2 kx2 .
Pzz + kz2 P = 0 , P = P (, kx , z) . (6.8)
The solution to the last equation is a simple exponential of the form
P (, kx , z) = Aeikz (,kx )z (6.9)
The wavefield recorded at the surface is denoted by p(t, x, z = 0) (the seismic data). In the (, kx , z) domain,
the wavefield is given by P (, kx , z = 0). The constant A can be evaluated by considering the wavefield at
z = 0 (the data).
P (, kx , z) = P (, kx , z = 0)eikz (,kx )z (6.10)
We have found the expression for the downward continuation operator. This is an operator the extrapolates
the wavefield at the surface into a depth z below the surface. The exploding reflector idea uses up-going
waves. This is why we will choose the minus sign in the extrapolator,
P (, kx , z) = P (, kx , z = 0)eikz (,kx )z (6.11)
r
2
kz = ( kx2 ) . (6.12)
v2
The story doesnt end here. It is clear that if we know the velocity v we can extrapolate our wavefield
recorded at z = 0 down into the earth, in other words, we can compute P (, kx , z) z the imaging condition
is obtained by transforming back to time and space using the inverse Fourier transforms,
P (, kx , z) Ft1 P (t, kx , z) Fx1 p(t, x, z). 4
Finally we apply the imaging condition:
I(x, z) = p(0, x, z)
Wavefield extrapolation can be also done in the t, x, z space. Note that wavefield extrapolation in (, kx ) is
equivalent to a convolution in the (, x) domain.
We have described a technique to migrate seismic data in the case where the velocity is laterally invariant,
v(z). This technique is called Gazdag migration.
The following MATLAB implementation permits to run constant velocity Gazdag migration; simple
modifications would allow us to migrate data in v(x, z) media.
6.1.2 Gazgad Migration: Examples
I have generated a syncline model using SeismicUnix. I have computed the associated zero offset section
and finally, I have migrated it using different velocities. In Figure (6.4) we portray the zero offset section
corresponding to the syncline model. In this example we are using a constant velocity medium (V =
1km/sec).
First, we migrate the zero-offset section using the correct velocity. The result is portrayed in Figure
(6.5). It is clear, that the migration process has removed diffraction patterns produced by multiple arrivals
of energy from the flanks of the syncline. In Figures (6.6) and (6.7) we have used the wrong velocity to
migrate the data.
4 F 1 is the inverse Fourier transform over t

t
function [dout] = down_constant_v(din,v,dx,dt);

% Migration using the phase shift method
% v is constant.
% din: input data
% v: velocity (MKS)
% dx: distance between receivers
% dt: sampling interval
% dout: migrated data
%
D = fft2(din) ; 2D fft of the data
nx = min(size(D));
nt = max(size(D));
% Compute wavenumebers, symmetries are imposed
knyq = pi/dx; dkx = 2*pi/(nx*dx); kx = [0:dkx:dkx*(nx/2) -(dkx*(nx/2-1):-dkx:dkx)];
wnyq = pi/dt; dw = 2*pi/(nt*dt); w = [0:dw:dw*(nt/2) -(dw*(nt/2-1):-dw:dw)];
i = sqrt(-1);
for l=1:nt
for j=1:nx
kzkz = w(l)*w(l)-v*v*kx(j)*kx(j); % Predicted squared vertical wavenumber
if kzkz > 0
C(l,j) = exp( i*sign(w(l))*dt*sqrt(kzkz));
else
C(l,j) = 0;
end
end
end
for l=1:nt
D = D.*C; % Downward continuation
IMAGE(l,:) = ones(1,nt)*D; % imaging condition
end
image = real(ifft(IMAGE,[],2));
dout = image;
Figure 6.3: Gazdag migration algorithm for constant velocity

Distance (km)
5
0
Time (sec)
Synthetic Data
Figure 6.4: Zero-offset synthetic data used to test Gazdag migration.
Midpoint (km)
5
0
Migrated Time (sec)
SUGAZMIG Migration Vmig=1.km/sec (Correct velocity)
Figure 6.5: Gazdag migration using the correct velocity.

Midpoint (km)
5
0
Migrated Time (sec)
SUGAZMIG Migration Vmig=.8km/s
Figure 6.6: Gazdag migration using the wrong velocity.
Midpoint (km)
5
0
Migrated Time (sec)
SUGAZMIG Migration Vmig=1.4km/s
Figure 6.7: Gazdag migration using the wrong velocity.

x (m) x10 4
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
0
2
t (s)
Figure 6.8: Zero offset synthetic data obtained from the SEG-EAEG sub-salt model
6.1.3 Phase shift Migration plus interpolation (PSPI)
We will study a modification to Gazdag migration that is capable of dealing with the problem of migrating
data in v(x, z).
The algorithm outlined in the previous section was valid for v(z). In this case, the algorithm is often
called Gazdag migration or phase-shift migration in v(z).
In PSPI the wavefield is downward continued using several velocities. The algorithm is equivalent to
running Gazdag migration for a set of reference velocities and then interpolating to compute the the local
wavefield by interpolating the wavefield obtained by propagating the data using reference velocities.
In Figure (6.8) I portrayed the zero offset data associated to the SEG-EAGE subsalt model. This model is
often used to study the performance of imaging algorithms. In Figure (6.9) the data has been migrated using
Gazgad v(z) algorithm, in other words the velocities of geological structure were assumed to be laterally
invariant. The result is portrayed in Figure(6.9). The correct velocities where used to run the PSMPI
algorithm, in this case 10 reference velocities per depth step were adopted. The result is displayed in Figure
(6.10). In general migration velocities are also called macro velocities or the macro model. These
velocities are usually a smooth representation of the real velocities of the subsurface. Techniques to estimate
migration velocities are often referred as migration velocity analysis.
x (m) x10 4
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0
0.2
0.4
z (m)
0.6
0.8
1.0
x10 4
Figure 6.9: Gazdag v(z) migration. Note that the incorrect assumption of laterally invariant velocity leads
to an image dominated by artifacts.
x (m) x10 4
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
0
0.2
0.4
z (m)
0.6
0.8
1.0
x10 4
Figure 6.10: PSMPI using 10 reference velocities per depth step. In this example the salt body is properly
migrated.
6.2 Summation along hyperbolic paths
We can describe the migration process in very simple terms by first understanding how to model seismic
data in constant velocity media.
Modeling: MODEL of the subsurface DATA
Imaging: DATA MODEL of the subsurface
When one uses the exploding reflector analogy, it is possible to imagine that a reflector is composed of
a continuous distribution of sources. To understand this concept, we will simulate a seismic reflection by
placing sources on a reflector and compute the total response as the superposition of individual sources. The
position of the source i in Figure (6.11) is denoted by xi , zi and the position of the receiver is given by xm
(this is also the midpoint position for our zero-offset experiment).
The traveltime for the ray going from the source i th to the receiver is
q
1
Ti = zi2 + (xm xi )2 (6.13)
v
If each source is represented by a time function s(t) (a wavelet), and if we assume that all the sources explode
at the same time (t = 0), the seismogram at the receiver xr is modeled as the superposition of individual
sources delayed in time according to the following equation
q
X 1
data(xm , t) = s(t zi2 + (xm xi )2 ) . (6.14)
i
v
This is a simple but very important equation that enables us to model seismic data (only primary reflections)5 .
It is clear that the reflector in Figure 6.11 needs an infinite number of sources very close to each other to be
properly modeled.
In Figure 6.12 we present a model that consists of 10 sources distributed in x z. Now, you can imagine
that this sources are small pieces a of broken reflector. The data generated by this model are a superposition
of hyperbolas. In the same figure I have also shown the migrated data, in this case the data has been
migrated with an algorithm that undo the process outline in equation (6.14). I will come back to this point
in a minute.
In Figure 6.13, I have used 20 sources to model the reflector, the data generated from this model and the
migrated image are shown in the same figure.
5 To simplify the problem, I didnt include amplitude terms, but bear in mind that there is an amplitude variation.
6.2. SUMMATION ALONG HYPERBOLIC PATHS 85
Xm
R
*
*
(xi,zi)
*
*
Reflector Reflector represented by
4 sources (*)
(xi,zi): position of source i
Figure 6.11: Reflectors are represented by a superposition of sources.
In Figure 6.14, I have used 200 sources to model the reflector, the data generated from this model and the
migrated image are shown in the same figure. In this case each piece of reflector in x z is an active source.
It is clear, that the exploding reflector model leads to a very simple integral formulation for modeling data
(equation (6.14). Now we have to find the equation to undo the modeling process (The migration algorithm).
Equation (6.14) is a forward operator (creates data from a subsurface model). The following equation
does the reverse process (creates a subsurface model from data)
X 1p 2
image(x, z) = d(xm , z + (xm x)2 ) . (6.15)
xm
v
The last equation that can be formally derived corresponds to a poor man Kirchoff migration algorithm.
Equation (6.15) permits to construct an image by summation along hypebolas in the t xm plane.
x (m)
0 200 400 600 800
0
200
z (m)
400
Subsurface model
Midpoint (m)
0 200 400 600 800
0
0.05
time (sec)
0.10
0.15
Data
x (m)
0 200 400 600 800
0
200
z (m)
400
Migrated data
Figure 6.12: Hyperbola summation for 10 sources

x (m)
0 200 400 600 800
0
200
z (m)
400
Subsurface model
Midpoint (m)
0 200 400 600 800
0
0.05
time (sec)
0.10
0.15
Data
x (m)
0 200 400 600 800
0
200
z (m)
400
Migrated data

x (m)
0 200 400 600 800
0
200
z (m)
400
Subsurface model
Midpoint (m)
0 200 400 600 800
0
0.05
time (sec)
0.10
0.15
Data
x (m)
0 200 400 600 800
0
200
z (m)
400
Migrated data

6.2.1 Subroutine F77 to do and undo wavefields
This is a prototype of a migration code in Fortran77. You have two options given by the variable adj,
adj=0 means modeling, adj=1 means undo the modeling (do the migration).
In this case, we have zero-offset data and a constant velocity medium.
subroutine hyperbola_sum(adj,vhalf,dt,dx,dz,nt,nx,nz,m,d)
c Zero offset example

c This code can be optimized.
c Input: m if adj=0 (Modeling) Output: d
c Input: d if adj=1 (Migration) Output: m
real m(500,200),d(500,200)
integer adj
if(adj.eq.0) then
do ix=1,nx
do it=1,nt
d(it,ix)=0.
enddo
enddo
endif
if(adj.ne.0) then
do ix=1,nx
do iz=1,nz
m(iz,ix)=0.
enddo
enddo
endif
do ir=1,nx ! loop on receivers
xr=(ir-1)*dx
do ix=1,nx ! loop on lateral position
x=(ix-1)*dx
do iz=1,nz ! loop on depth
z=(iz-1)*dz
t=(sqrt(z*z+(xr-x)**2))/vhalf
it=1.5+t/dt
if(it.le.nt) then
if(adj.eq.0) d(it,ir)=d(it,ir)+m(iz,ix)
if(adj.ne.0) m(iz,ix)=m(iz,ix)+d(it,ir)
endif
enddo
enddo
enddo
return
end

Seismic Introduction

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Seismic Introduction

Enviado por

Direitos autorais:

Formatos disponíveis

Introduction to Seismic Imaging

(Refraction and Reflection Seismology)

Figure 1.1: Plane waves and spherical waves.

1.1 Snells Law

from the above equation we can recognize the following

Replacing sin(i) and sin(j) into equation (1.5)

the latter is called Snells law, which I will re-write as follows:

P =(xp,zp); Q =(xq,zq); o = (x,0)

Ray paths Traveltimes

Ray paths Traveltimes

and for the reflected wave,

if v2 > v1 => j > i . (1.12)

When can adopt i as the variable of our problem, and write

which can be written as follows

sin(i)/v1 = sin(i)/v1 ==> i =i

Figure 1.6: A P wave incident on an interface between contrasting materials.

Reflected Wave Direct Wave

Figure 1.7: Rays in a one layer structure over a half space.

1.1.1 T X refraction profiles

First, we will analyze the direct wave:

The total travel-time for a direct wave traveling from A to D is given by

Td (x) = x/v1 . (1.16)

It is clear from the diagram that AB = CD, then

Using the critical angle condition sin(ic ) = v1 /v2 we can write

Figure 1.8: Geometry of a refraction profile.

1. From T (x) measure the slopes and compute v1 and v2

2. Compute ic using the expression sin(ic ) = v1 /v2

3. Measure t1 and compute z = (v1 t1 )/(2 cos(ic ))

1.1.2 T X Refraction curves for a dipping layer

Figure 1.9: Travel-time curve for a refraction profile.

(the dip of the layer).

The total travel-time for the Experiment I (Figure 3.10a) is given by

BC = x cos() Z tan(ic ) z tan(ic ) . (1.26)

The time in terms of our variables and unknowns is given by

1 tan(ic ) 1 tan(ic ) x cos()

the last equation can be reduced by introducing the following equations

After substituting the last two equation in (3.27) we obtain

The last expression corresponds to a straight line with slope:

and intercept time (T (x = 0))

The direct ray can be used to obtained v1 .

which also defines a straight line in T X,

4. Use the intercept times to compute z and Z.

1.1.3 Fitting straight lines to first arrival picks

event : Direct Wave pick x(m) t(msec)

event : Refracted wave pick x(m) t(msec)

Figure 1.11: Refraction experiment. Travel-time curves

The minimization of l1 leads to the unknowns of our problem a and b.

Least squares fitting

Figure 1.12: Least-squares curve fitting.

Figure 1.13: Least-squares curve fitting in the presence of outliers.

1.2.1 T X curves for a primary reflection

The time as a function of the source receiver distance x becomes:

We now define the two-way zero offset travel time

The last equation can be rearranged in the following form:

which is the equation of a hyperbola1 in T x.

We can use the cosine theorem to write

= 90o + => cos(90o + ) = sin() . (1.47)

The travel-time becomes

We can re-write the latter as

Figure 1.14: A primary reflection. The horizontal layer case.

Figure 1.15: Common Shot Gather(CSG).

Figure 1.16: A primary reflection. The dipping interface case.

Head wave: Th = 2 z cos(i

We can start by taking derivatives of Tr